Seedance 2.5 Reference to Video
bytedance/seedance/2.5/reference-to-videoByteDance's next-generation reference-to-video model, generating video from multimodal references (images, videos, audio) and locking a character, set, and palette across a full take up to 30 seconds for production-grade consistency.
- Base price
- $1.12 / run
- Execution
- async
- Model type
- video
- Input fields
- 8
Try the model
Playground
PNG, JPEG, WebP, or GIF · 20 MiB maximum each
Your output will appear here
Complete the inputs, then click Run.
Specifications
Pricing
- Base price
- $1.12 / run
- Billing formula
- (usage.total_seconds ?? ((params.duration ?? 5) == 'auto' ? 5 : (params.duration ?? 5))) * ((params.resolution ?? '720p') == '480p' ? 0.099703264095 : (params.resolution ?? '720p') == '720p' ? 0.224332344214 : (params.resolution ?? '720p') == '1080p' ? 0.555222551929 : (0 / 0)) + (params.has_input_videos ? usage.input_video_seconds : 0) * ((params.resolution ?? '720p') == '480p' ? 0.059821958457 : (params.resolution ?? '720p') == '720p' ? 0.134599406528 : (params.resolution ?? '720p') == '1080p' ? 0.331691394659 : (0 / 0))
Context & modalities
- Input
- Schema-defined
- Output
- video
Capabilities
- Chat
- Not supported
- Vision
- Not supported
- Reasoning
- Not supported
- Structured output
- Not supported
- Function calling
- Not supported
- Audio input
- Not supported
Access
- Provider
- Bytedance
- Model ID
- bytedance/seedance/2.5/reference-to-video
- Execution
- async
- API
- Unified Run API
- Endpoint
- /v1/run
API README
Seedance 2.5 Reference to Video
ByteDance's next-generation multi-reference video model — combines reference images, videos, and audio to lock a character, set, and palette across a full take up to 30 seconds for production-grade consistency.
- Need text-only generation? Try Seedance 2.5 Text to Video.
- Need single-image driven generation? Try Seedance 2.5 Image to Video.
Highlights
Multi-modal references — Combine reference images, videos, and audio in a single request. Refer to them in the prompt as @Image1, @Video1, @Audio1, etc. Up to 50 files across all modalities.
Native audio generation — Automatically creates synchronized sound effects, ambient audio, and lip-synced speech. Toggle with generate_audio parameter.
Style + identity lock — Use references to hold visual style, motion patterns, and subject identity steady across the whole clip.
Flexible duration — Generate 4 to 30 second clips in a single request.
Pricing
Rates use Volcengine's non-promotional list price converted at 1 USD = 6.74 CNY.
| Resolution | Output price per second | Input-video price per second |
|---|---|---|
| 480p | $0.10 | $0.06 |
| 720p | $0.22 | $0.13 |
| 1080p | $0.56 | $0.33 |
Default: 5 output seconds at 720p = $1.12. Video references add the input-video rate using the server-probed trusted aggregate input duration. Only the exact Seedance 2.0 standard, Fast, and Mini reference-to-video modes and Seedance 2.5 reference-to-video mode, each with non-empty videos, explicitly use the isolated lenient-duration policy; the default media probe, Webhook validation, empty-video requests, and every other model or caller remain strict. Every legal domain in this isolated policy uses the system resolver, standard HTTP transport, and ProxyFromEnvironment without all-address public-DNS validation, IP pinning, or DNS-rebinding protection. Each redirect hop repeats the base URL and redirect validation. Invalid URL or host, redirect-policy, count-limit, per-item or aggregate size-limit, parser-resource-limit, and successfully parsed aggregate duration-limit failures fail closed. All other fixed failures—dns_unavailable, connect_error, network_timeout, tls_error, http_status, download_error, empty_body, mime_mismatch, magic_mismatch, unsupported_media, malformed_media, and duration_unavailable—use the 30-second family cap for the whole request once, not per video. Logs and errors expose only fixed stage/reason, video index/count, host hash, bytes, elapsed time, fallback marker, and policy version; complete URLs, queries, signatures, response bodies, and media content remain redacted. New requests freeze trusted-probe policy v4; existing v3, v2, and legacy predictions retain their frozen settlement semantics. This isolated application policy requires no new server, Kubernetes, proxy, egress, or environment configuration. Image and audio references do not add this video-input charge. Audio generation is included at no extra cost.
When to Use
| ✅ Good fit | ❌ Consider alternatives |
|---|---|
| Brand-consistent video production | Real-time video streaming |
| Style-guided content creation | Long-form video (>30s) |
| Video-to-video style transfer | Simple text prompts (use T2V) |
| Audio-synced video generation | Single image animation (use I2V) |
| Multi-reference scene composition | Quick prototyping without assets |
Input Requirements
- Reference images/videos/audio as publicly accessible URLs
- Supported image formats: JPG, PNG, WebP (max 30 MB, up to 30 images)
- Supported video formats: MP4, MOV (up to 10 videos, 1.8–30.2s each, combined ≤ 30.2s)
- Supported audio formats: MP3, WAV (up to 10 files, combined ≤ 30.2s)
- Audio cannot be used alone — must pair with at least one reference image or video
Technical Specs
| Spec | Value |
|---|---|
| Input | Text + reference images/videos/audio |
| Output | MP4 video with optional audio |
| Resolution | 480p / 720p / 1080p |
| Aspect ratios | 16:9, 9:16, 1:1, 4:3, 3:4, 21:9 |
| Duration | 4–30 seconds |
| Execution | Async (submit → poll for result) |
| Typical latency | 2–10 minutes |
Related
- Seedance 2.5 Text to Video — Generate video from text prompts
- Seedance 2.5 Image to Video — Single image to video
Start building
Send your first request
OpenAI-compatible endpoint with unified authentication and usage tracking.
# Preserve Unicode and shell metacharacters in the request body.
result=$(printf '%s' 'ewogICJtb2RlbCI6ICJieXRlZGFuY2Uvc2VlZGFuY2UvMi41L3JlZmVyZW5jZS10by12aWRlbyIsCiAgImltYWdlcyI6IFsKICAgICJodHRwczovL3N0YXRpYy5zYW5kYmFzZS5haS9leGFtcGxlcy9ieXRlZGFuY2Uvc2VlZGFuY2UvMi4wL3JlZmVyZW5jZS10by12aWRlby9pbnB1dF9pbWFnZV9zYWZlX29jZWFuXzAuanBnIgogIF0sCiAgInByb21wdCI6ICJVc2UgSW1hZ2UxIGFzIHRoZSB2aXN1YWwgcmVmZXJlbmNlIGFuZCBWaWRlbzEgYXMgdGhlIG1vdGlvbiByZWZlcmVuY2UuIENyZWF0ZSBhIHBlYWNlZnVsIGNpbmVtYXRpYyBzaG9yZWxpbmUgc2NlbmUgd2l0aCByb2xsaW5nIG9jZWFuIHdhdmVzIGF0IGdvbGRlbiBob3VyLCBzb2Z0IGZvYW0sIG5hdHVyYWwgd2F0ZXIgbW92ZW1lbnQsIGFuZCBubyBwZW9wbGUuIiwKICAidmlkZW9zIjogWwogICAgImh0dHBzOi8vc3RhdGljLnNhbmRiYXNlLmFpL2V4YW1wbGVzL3BpeHZlcnNlL3NvdW5kLWVmZmVjdHMvaW5wdXRfdmlkZW9fMC5tcDQiCiAgXSwKICAiZHVyYXRpb24iOiA1LAogICJyZXNvbHV0aW9uIjogIjcyMHAiLAogICJnZW5lcmF0ZV9hdWRpbyI6IHRydWUKfQ==' \
| base64 --decode \
| curl --fail-with-body --silent \
-X POST "https://api.sandbase.ai/v1/run" \
-H "Authorization: Bearer $SANDBASE_API_KEY" \
-H "Content-Type: application/json" \
--data-binary @-)
run_id=$(printf '%s' "$result" | jq -r .id)
for attempt in $(seq 1 120); do
status=$(printf '%s' "$result" | jq -r .status)
case "$status" in completed|failed|timeout) break ;; esac
sleep 2
result=$(curl --fail-with-body --silent \
-H "Authorization: Bearer $SANDBASE_API_KEY" \
"https://api.sandbase.ai/v1/run/$run_id")
done
status=$(printf '%s' "$result" | jq -r .status)
[ "$status" = completed ] || { echo "Generation ended: $status" >&2; exit 1; }
printf '%s\n' "$result"Choose your model
Compare models
| Model | Context | Input / 1M | Output / 1M | Released |
|---|---|---|---|---|
Seedance 2.5 Reference to VideoThis model Bytedance | — | — | — | Jul 1, 2026 |
Bytedance | — | — | — | Jul 1, 2026 |
Bytedance | — | — | — | Jul 1, 2026 |
Bytedance | — | — | — | Jun 16, 2025 |
Bytedance | — | — | — | Jun 16, 2025 |
Bytedance | — | — | — | Dec 23, 2025 |